AI search & GEO
AI Voice Generators Compared: How to Make Voiceovers for Videos and Ads
Which tool lets you use the voice in an ad without paying, how many words fit in 30 seconds, what happened to PlayHT and what the law says before you clone a voice. All checked on official pages, with dates.
An AI voice generator turns a script into a natural-sounding voiceover in seconds: you type the text, pick a voice and download the audio for your video, ad or podcast intro. Generating it is the easy part. The hard part is picking a tool whose license lets you run it in an ad.
We checked the official pricing and terms pages of 12 tools in late October 2026. Three ban commercial use on their free plan, four allow it (two with fine print), and one that many lists still recommend, PlayHT, no longer works. «ai voice generator» gets about 74,000 searches a month in the U.S. (DataForSEO, October 2026), so you're not the only one confused.
Here's the comparison, the steps, how many words fit in 30 seconds and the law on voice cloning. It's part of our AI search and GEO guides.
On this page
What is an AI voice generator?
An AI voice generator is text-to-speech software that uses AI models to read text in a voice that sounds human.
An AI voice generator creates audio from written text using a speech model. You pick a voice from a library, or clone your own, and the tool reads your script with intonation, pauses and the language you choose. Marketers use it for video voiceovers, ads, radio spots, tutorials and podcast intros.
What it isn't: a voice actor who interprets your brand. It won't know where the emphasis goes unless you show it. If you want a show with personality, read how to start a podcast first and decide which parts need a human voice.
Best AI voice generator for ads: 10 tools compared
The best AI voice generator for an ad is the one whose license covers commercial use and whose accent matches your audience.
Checked in late October 2026, on each tool's pricing and terms pages:
| Tool | Free plan | Commercial use on free | Price from |
|---|---|---|---|
| ElevenLabs | 10,000 credits a month | No; must credit ElevenLabs | $6 a month |
| Adobe Firefly | Limited free uses | Yes, with its Firefly Speech model | $9.99 a month |
| Google Cloud Text-to-Speech | 1 million characters a month on HD voices | Pay as you go, no commercial limit | $30 per million characters |
| Microsoft Azure Speech | 500,000 characters a month | Pricing page doesn't restrict it | Not listed |
| CapCut | Free on the web | Yes, with the «Commercial License» filter | Free |
| Canva | A selection of free voices | Not stated on the page | Not stated |
| Murf | No free Studio plan | Yes, on paid plans | $19 a month, billed yearly |
| Speechify Studio | 600 credits | No | $100 a year |
| TTSMaker | 20,000 characters a week | Yes, free | Free |
| SpeechGen | 3,000 characters a day for 7 days | Yes, on every plan | $4.99 |
Two more notes. OpenAI's text-to-speech API requires you to tell listeners the voice is AI-generated. And if you also run ads in Spanish, ElevenLabs separates Spain from Latin American accents, which most tools don't.
What happened to PlayHT?
PlayHT is gone: its website no longer resolves, and secondary sources report it shut down after Meta acquired its parent company in 2025. Some top-ranking lists still recommend it. If yours does, the list is out of date.
Free AI voice generator options you can use in an ad
A free AI voice generator only works for an ad if its license allows commercial use; the monthly quota rarely matters.
We wrote our own 30-second bakery spot: 60 words, 344 characters. Here's how many reads of that script each free plan gives you:
| Free plan | Reads of our script | OK for an ad? |
|---|---|---|
| ElevenLabs | 29 a month | No |
| TTSMaker | 58 a week | Yes |
| SpeechGen | 8 a day | Yes |
| Azure (F0) | 1,453 a month | Yes, via API |
| Google Chirp 3 HD | 2,906 a month | Yes, via API with billing on |
If you searched for «ai voice generator free» or «text to speech free» for something you'll publish, the quota is plenty almost everywhere. The license is what locks you out.
How to make an AI voiceover for a video
To make an AI voiceover, write the script to length, pick a voice with commercial rights and mix the audio in your editor.
An AI voice over for a 30-second ad takes an afternoon:
-
Measure the slot
Note how many seconds the video or spot runs. That's your script's limit.
-
Write the script to length
At 150 words per minute, 30 seconds is about 75 words. Count them with our word counter, which shows speaking time too.
-
Write it the way it's said
Numbers in words, no abbreviations, brand names spelled the way they sound: «Twelve Main Street», not «12 Main St.».
-
Pick a voice with commercial rights
Use the table. On CapCut, switch on the «Commercial License» filter.
-
Check the accent
Listen to a full sentence before you commit. «English» or «Spanish» in a menu says nothing about region.
-
Generate two or three versions
Change the voice or the speed and keep the one that sounds like your brand.
-
Mix it and label it if needed
Duck the music under the voice, and label the video as AI if it imitates a real person.
If an AI writes the script, give it the length, audience and tone. That's prompt engineering applied to audio:
Write a 30-second radio spot (70 words max) for [business] in [city]. Audience: [who]. Goal: [what you want them to do]. Facts: [offer, address, hours]. Write it to be read aloud: short sentences, numbers spelled out, no abbreviations or symbols, brand name at the start and the end. Friendly American English.
Words per minute: how long your script runs
Voiceover runs at about 150 words per minute: 50 words in 20 seconds, 75 in 30 and 150 in 60.
That's the average rate of conversational speech in the U.S., the figure Voices.com cites from the National Center for Voice and Speech, and the default in our words per minute calculator, the word counter above.
| Length | At 150 words per minute | At 130 (slow read) |
|---|---|---|
| 15 seconds | 38 words | 33 words |
| 20 seconds | 50 words | 43 words |
| 30 seconds | 75 words | 65 words |
| 60 seconds | 150 words | 130 words |
How many words is a 30-second ad?
About 75 at a normal pace, 65 if you want room to breathe. Our spot has 60 words and runs 24 seconds. Translating? The same ad needed 71 words in Spanish and 72 in Portuguese, so trim before you localize.
AI voice changer: turn your own recording into another voice
An AI voice changer turns a recording of you into a different voice and keeps your timing and delivery. It doesn't read a script. It copies your performance.
Most people searching for this want a live voice changer for games or Discord. That's a different product category and we didn't test it. This section is for marketers who want a better voice on a recorded spot.
It's the fix when text to speech sounds flat. Record the script on your phone, stress the words that matter and let the tool swap the timbre. ElevenLabs' documentation says the output follows your cadence and tone, whispers and laughs included. It also keeps the accent of your recording: read it with an American accent and it stays American, whatever voice you pick.
What ElevenLabs lists:
- Length: up to 5 minutes and 50 MB per file. For a longer video, cut it into scenes.
- Cost: 1,000 credits per minute on its pricing page. A 30-second spot burns 500, so Starter's 30,000 credits cover about 60 versions. Starter includes commercial use.
- Formats: audio (MP3, WAV, M4A) and video (MP4, MOV), plus a background-noise remover.
Need twenty versions of one script? Use text to speech. Want one take with your own timing? Record it and convert it. If you pick the voice of a real person, the consent rules in the next section apply.
AI dubbing: how to dub a video into other languages
AI dubbing translates a video's audio into another language and generates the translated voice, often in the speaker's own voice. Two of the three routes below cost nothing.
- YouTube: per its help page, automatic dubbing is on by default for eligible channels. An English video can be dubbed into 20 languages, Spanish and Portuguese among them. It skips videos over 120 minutes or with little speech, you can't edit the dub, and you can choose to publish manually to review it first. Dubbed videos are marked "auto-dubbed" in the description.
- Instagram and Facebook Reels: Meta says translating reels with Meta AI is free. It imitates your voice, offers optional lip sync and labels each reel as translated. It's open to public Instagram accounts in countries with Meta AI and to Facebook creators with 1,000 followers or more.
- ElevenLabs: 90+ languages, files up to 180 minutes and 2 GB, and each speaker keeps their voice. Its pricing page lists 3,000 credits per minute for automatic dubbing without a watermark, and each target language is billed separately. A 45-second ad into two languages costs 4,500 credits, 15% of Starter's 30,000. Free-plan dubs carry a watermark.
Watch the clock. Our spot took 60 words in English and 71 in Spanish, 18% more. If the dub runs longer than the picture, it spills out of the slot. Before you publish, have a native speaker check the slogans: a literal translation rarely sounds like an ad.
AI voice cloning: consent, the ELVIS Act and the NO FAKES Act
Voice cloning is legal with your own voice or with clear consent; using someone else's voice to sell something without permission is not.
- Robocalls: in a 2024 ruling, the FCC confirmed that AI-generated voices count as «artificial» under the TCPA, so automated calls with them need prior express consent.
- Tennessee: the ELVIS Act, signed in March 2024, added voice, including AI-simulated voice, to the state's personal rights law.
- California: AB 2602 limits contract clauses that let a company replace a performer with a digital replica of their voice, for performances fixed since January 1, 2025.
- Federal: the NO FAKES Act is not law. It cleared the Senate Judiciary Committee in June 2026 and has stalled since.
- The tools: ElevenLabs bans cloning a living person without explicit, informed consent, and its professional clone only works with your own voice.
Do you have to label AI voices on YouTube or Instagram?
If it imitates a real person, yes. YouTube's disclosure rules cover realistic content that makes a real person appear to say something they didn't, but cloning your own voice for voiceovers is exempt. Meta requires its AI label on realistic-sounding audio. TikTok asks you to label realistic AI content.
Frequently asked questions about AI voice generators
Is ElevenLabs free?
Yes, with limits that matter for marketers. The free plan gives 10,000 credits a month, but it doesn't include commercial use, and anything you publish has to credit ElevenLabs in the title. That rules it out for ads and client work. On ElevenLabs pricing, the Starter plan is $6 a month ($5 billed yearly) and adds commercial rights and instant voice cloning. Check the current plan page before you buy.
What is the best AI voice generator?
It depends on the job. For realism and the widest accent range, ElevenLabs on a paid plan. For high volume through an API, Google Cloud or Microsoft Azure voices. For free commercial use, TTSMaker or SpeechGen. For editing the video in the same place, CapCut. There's no universal winner: run your actual script through two or three voices and keep the one that sounds like your brand.
How do I clone my voice with AI?
Record one or two minutes of clean speech in a quiet room, no music, then upload it to a tool that offers cloning, such as ElevenLabs from its Starter plan. For its professional clone, ElevenLabs makes you verify it's your own voice by reading prompts aloud. Never clone someone else's voice without explicit, written consent, and if it's an employee's voice, get that consent before you record.
How much does an AI voice generator cost?
From $0 to a few dollars a month for most small businesses. Free plans exist, but several ban commercial use. Paid plans we checked start at $6 a month (ElevenLabs), $9.99 (Adobe Firefly) and $19 billed yearly (Murf). API voices bill per character: Google charges $30 per million characters for its Chirp 3 HD voices after the monthly free tier. A 30-second ad is about 350 characters.
What are the ethical considerations of using AI voice generators?
Three big ones: consent, disclosure and honesty. Consent, because a voice belongs to a person and cloning it without permission can break the law. Disclosure, because listeners have a right to know when a realistic voice is synthetic, and YouTube and Meta require labels in some cases. Honesty, because a fake testimonial read by an AI voice is still a fake testimonial. Use AI voices to narrate, not to impersonate.
What are the limitations of AI voice generators?
They read, they don't perform. AI voices still stumble on brand names, acronyms and numbers, can sound flat in long scripts and don't know where the emphasis goes unless you mark it with punctuation. Accents are hit or miss: many menus say «Spanish» or «Portuguese» without a country. And licenses vary more than quality does, so the real limit is often legal, not technical.
How does an AI voice generator work?
A neural speech model reads your text, predicts the rhythm, stress and sound of each sentence, and generates the audio. It learned that from hours of recorded human speech, so it picks up pauses and intonation, not just pronunciation. That's why punctuation changes the delivery. Voice cloning adds one step: a short sample of a speaker steers the model toward that person's timbre. Cleaner samples give cleaner clones.
Keep going with these guides
- AI search & GEO Prompt Engineering for Marketers: How to Write Prompts That Work (Tested in 4 AIs) Read the guide
- AI search & GEO What Is a Chatbot? How to Add One to Your Website or WhatsApp Read the guide
- AI search & GEO AI Browsers: What They Are and What They Mean for Your Website Read the guide